AASS Seminar
19 mars 2026 13:00 – 14:00 Hörsal T, Teknikhuset
The research centre AASS arranges a seminar with Zahra Alijani and Vojtech Molek from Ostrava University.
Who: Zahra Alijani and Vojtech Molek from Ostrava University
What: Seminar: "Fractional Calculus in Deep Learning: Unified Perspectives on Activation and Loss Functions"
When: March 19, 2026 starting at 13:00.
Where: Hörsal T, Teknikhuset
Remote attendance: https://oru se.zoom.us/j/65096623293
Abstract
This seminar explores the integration of fractional calculus into deep learning through both activation and loss function design.
First, we examine fractional activation functions—such as fractional GELU, Mish, and sigmoid—derived via Grunwald–Letnikov operators, highlighting their flexibility and computational challenges. Second, we introduce fractional variants of classical loss functions, including fractional MSE, Huber, and cross-entropy, which enable smooth control of error penalization. Empirical results across vision benchmarks show that fractional methods can offer performance benefits but often incur substantial computational cost. Together, these findings provide a unified view of fractional concepts in neural network learning dynamics.